Spunlace nonwoven fabric rewinding equipment collaborative control method and system based on internet of things

Abstract

The application discloses a spunlace non-woven fabric rewinding equipment collaborative control method and system based on the Internet of Things, and relates to the field of electric digital data processing.The method comprises the following steps: obtaining historical data and preprocessing to obtain a plurality of three-dimensional data sequence sets of roll diameter-torque-tension under normal working conditions; dividing the plurality of three-dimensional data sequence sets into a plurality of roll diameter slice sample groups and screening to obtain a non-woven fabric flat data set; performing nonlinear fitting on the non-woven fabric flat data set and constructing non-woven fabric flat upper and lower tracks; based on the non-woven fabric flat upper and lower tracks, real-time marking of an abnormal data interval in the current rewinding process is performed, and abnormality determination is performed in cooperation with an image detection mechanism; after the current rewinding process is completed, the non-woven fabric flat upper and lower tracks are updated and optimized. The application can dynamically adapt to roll diameter changes, accurately mark abnormal intervals and accurately determine abnormal working conditions in cooperation with an image detection mechanism, thereby effectively improving the quality control level of spunlace non-woven fabric rewinding production.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of digital data processing, in particular to a spunlace nonwoven fabric rewinding equipment collaborative control method and system based on Internet of Things. BACKGROUND

[0002] The spunlace nonwoven fabric has the characteristics of light and thin texture and high tensile sensitivity, and the tension stability is strict during the rewinding process. Tension fluctuation easily leads to abnormal working conditions such as deviation and wrinkle. In the field of medical and health products, the rewinding flatness of spunlace nonwoven fabric is a key indicator to measure product quality. During the rewinding process, the torque fluctuation of the unwinding mechanism is the core inducement of uneven tension, which easily leads to quality defects such as deviation and wrinkle of nonwoven fabric. With the popularization of Internet of Things technology in industrial production, it is possible to collect equipment operation data in real time, but the existing control method of rewinding equipment has obvious shortcomings.

[0003] Firstly, the existing control method of rewinding equipment does not consider the influence of roll diameter change on torque reference. Under different roll diameters of nonwoven fabric parent roll, due to the difference in physical characteristics of moment of inertia and friction, the reference torque required to maintain uniform tension is not the same. The existing method mostly uses fixed threshold or static model, and cannot dynamically adjust the torque reference according to the roll diameter, resulting in low accuracy of abnormal judgment under high roll diameter and low roll diameter conditions. At the same time, the difference in torque fluctuation sensitivity caused by roll diameter is ignored. The influence of torque fluctuation on tension is different under different roll diameters, and the influence of torque fluctuation on tension is more significant at large roll diameter. The traditional control method does not distinguish the fluctuation sensitivity under different roll diameters, resulting in frequent misjudgment and omission.

[0004] Moreover, the existing control method of rewinding equipment is not reasonable in allocating image processing resources. In the medium and high speed rewinding scene, the nonwoven fabric moves fast, if the image detection mechanism continuously performs full-time image recognition without distinction, high-frequency image acquisition and edge detection processing are required, which will occupy a large amount of computing resources, easily leading to judgment delay and missing the best alarm and correction opportunity. However, the existing control method lacks a pre-marking mechanism for electrical data abnormal interval, and it is difficult to realize efficient collaboration with the image detection mechanism. SUMMARY

[0005] In order to accurately identify torque abnormalities, reduce the misjudgment rate of nonwoven fabric deviation abnormalities, improve the quality control level of spunlace nonwoven fabric rewinding production, realize the dynamic self-adaptation of the control mechanism to the change of roll diameter, and solve the problems of lack of pre-marking mechanism for electrical data abnormal interval in the existing rewinding equipment control method, difficulty in efficient collaboration with the image detection mechanism, and unreasonable allocation of image processing resources, the present application provides a spunlace nonwoven fabric rewinding equipment collaborative control method and system based on Internet of Things, and the technical scheme is as follows:

[0006] In a first aspect, the present application provides a spunlace nonwoven fabric rewinding equipment collaborative control method based on the Internet of Things, which comprises the following steps: obtaining historical time series data of multiple rewinding processes and preprocessing, extracting historical time series data under normal working conditions, and obtaining a plurality of three-dimensional data sequence sets of roll diameter-torque-tension; constructing a roll diameter window mechanism to divide the plurality of three-dimensional data sequence sets into a plurality of roll diameter slice sample groups, determining a tension uniformity threshold to filter the roll diameter slice samples, and obtaining a nonwoven fabric flatness data set; based on the nonwoven fabric flatness data set, calculating the sample weight of each sample in the data set and performing nonlinear fitting to obtain a fitting curve under ideal working conditions; based on the fitting curve under ideal working conditions, constructing nonwoven fabric flatness upper and lower tracks; based on the nonwoven fabric flatness upper and lower tracks, real-time marking the abnormal data interval in the current rewinding process, and cooperating with the image detection mechanism in the rewinding equipment to perform abnormality judgment; after the current rewinding process is completed, the range interval of the nonwoven fabric flatness upper and lower tracks is optimized and updated.

[0007] Among them, the nonwoven fabric flatness data set is filtered based on the tension uniformity threshold to obtain the fitting curve under the ideal working condition, the fitting curve is a curve function between the roll diameter and the torque, and the roll diameter is the independent variable and the torque is the dependent variable. The nonwoven fabric flatness data set is further filtered to obtain a set of core data points, so as to calculate the torque standard deviation of each roll diameter slice sample group, and thus based on the fitting curve under the ideal working condition and the three-sigma law, the nonwoven fabric flatness upper and lower tracks corresponding to each roll diameter slice sample group are obtained as the abnormality detection standard.

[0008] Preferably, the sensing device is deployed to synchronously collect roll diameter data sequence of the nonwoven fabric parent roll, torque data sequence of the unwinding mechanism, and tension data sequence of the nonwoven fabric in real time, obtain three-dimensional data of multiple nonwoven fabric complete rewinding processes as historical time series data, and normalize all historical time series data. Based on the original abnormality detection method, the normalized historical time series data is divided into normal working conditions or abnormal working conditions, the normalized historical time series data under normal working conditions in each rewinding process is extracted, and a three-dimensional data sequence set of roll diameter-torque-tension in each rewinding process is obtained.

[0009] Preferably, the roll diameter window reference length is set according to the quality parameters of the spunlace nonwoven fabric and the operating speed of the rewinding equipment in the actual application scenario. The normalized roll diameter value range is divided into multiple roll diameter intervals of equal length. The three-dimensional data sequence set of each rewinding process is divided according to the roll diameter window reference length, resulting in several roll diameter slice samples corresponding to each rewinding process. Roll diameter slice samples of the same roll diameter interval in each rewinding process are grouped into a roll diameter slice sample group, resulting in several roll diameter slice sample groups. The number of roll diameter slice sample groups is equal to the reciprocal of the roll diameter window reference length. The tension data sequence in each roll diameter slice sample within each roll diameter slice sample group is extracted and divided into four parts. The interquartile method is used to calculate the interquartile range of each tension data sequence as the tension fluctuation of the corresponding roll diameter slice sample, thus obtaining the tension fluctuation of each roll diameter slice sample within each roll diameter slice sample group. Based on the tension fluctuation, a histogram corresponding to each roll diameter slice sample group is constructed, and the maximum inter-class variance method is used to determine the tension uniformity threshold of each roll diameter slice sample group. Roll diameter slice samples within each roll diameter slice sample group whose tension fluctuation is less than the corresponding tension uniformity threshold are taken as tension uniform samples. The corresponding roll diameter-torque data pairs in the tension uniform samples are taken as data points, and the set of all data points in all tension uniform samples is taken as the nonwoven fabric flattening dataset.

[0010] Preferably, the tension fluctuation of each roll diameter slice sample within each roll diameter slice sample group is extracted, and based on the nonwoven fabric flatness dataset, the tension fluctuation of each tension uniform sample within each roll diameter slice sample group is extracted. The mean of the tension fluctuation of each roll diameter slice sample within each roll diameter slice sample group is used as the benchmark fluctuation value of the corresponding roll diameter slice sample group. The ratio between the tension fluctuation of each tension uniform sample within each roll diameter slice sample group and the corresponding benchmark fluctuation value is calculated. The natural exponential function is used to perform inverse mapping of the comparison value, and the mapped value is used as the sample weight of the corresponding tension uniform sample to obtain the sample weight of each tension uniform sample within each roll diameter slice sample group. A mature nonlinear fitting tool in the field of industrial control is directly called, and the nonwoven fabric flatness dataset is used as input. During the iterative optimization process of the fitting curve function, the sample weight of each tension uniform sample is introduced to calculate the weighted loss until the fitting curve function with the minimum weighted loss is obtained as the fitting curve under ideal working conditions.

[0011] Preferably, the absolute value of the vertical distance from the data point to the fitted curve under ideal working conditions is taken as the torque deviation distance of the corresponding data point. The torque deviation distance of each data point in the nonwoven fabric flattening dataset is calculated to obtain a set of deviation distances. Statistical calculations are performed on the set of deviation distances to obtain the median and interquartile range values. A radius threshold is set, with its upper limit being the sum of the median and interquartile range values ​​of the deviation distance set and its lower limit being the median value. The upper limit of the radius threshold is used as the initial value, and a step size for gradually decreasing the radius threshold is set based on the upper limit of the radius threshold. The set of values ​​for the radius threshold is obtained by taking the lower limit of the value; the interquartile range of the deviation distance set of 0.1 times is used as the threshold neighborhood coefficient, the sum of the radius threshold and the threshold neighborhood coefficient is used as the upper limit of classification, and the difference between the two is used as the lower limit of classification. Based on the torque deviation distance, the data points in the non-woven fabric flattening dataset are classified and filtered. Data points with a torque deviation distance less than the lower limit of classification are classified as core data points, data points with a torque deviation distance greater than the upper limit of classification are classified as edge data points, and data points with a torque deviation distance between the upper limit of classification and the lower limit of classification are classified as fuzzy data points.

[0012] Preferably, a classification loss function is constructed, setting the classification loss function value of fuzzy data points to 1 and the classification loss function value of core data points and edge data points to 0; based on a certain value within the radius threshold value set, fuzzy data points under that value are statistically analyzed, and based on the sample weights of the corresponding samples in the nonwoven fabric flattening dataset, a corresponding sample weight is matched for each fuzzy data point under that value, and the cumulative value of the product of the classification loss function value of all fuzzy data points under that value and the corresponding sample weight is used as the weighted classification loss value corresponding to that value; the set of radius threshold values ​​is traversed to obtain the weighted classification loss value corresponding to each radius threshold value in the set, and the radius threshold value corresponding to the minimum value among all weighted classification loss values ​​is used as the optimal radius threshold, thus obtaining the corresponding set of core data points.

[0013] Preferably, the core data points are categorized based on the roll diameter range of each roll diameter slice sample group, and the torque data of each core data point within each roll diameter slice sample group is extracted. The torque standard deviation of the core data points within each roll diameter slice sample group is calculated. Based on the three Sigma law, the sum of the fitted curve function under ideal working conditions and three times the torque standard deviation of a certain roll diameter slice sample group is taken as the nonwoven fabric flattening upper rail corresponding to that roll diameter slice sample group. The difference between the fitted curve function under ideal working conditions and three times the torque standard deviation of a certain roll diameter slice sample group is taken as the nonwoven fabric flattening lower rail corresponding to that roll diameter slice sample group. Similarly, the nonwoven fabric flattening upper and lower rails corresponding to each roll diameter slice sample group are obtained.

[0014] Preferably, the three-dimensional data of the current rewinding process is collected in real time and normalized. The normalized roll diameter-torque data pair is used as the current data point. Based on the normalized roll diameter data at the current moment, the current data point is classified into the corresponding roll diameter slice sample group. The corresponding nonwoven fabric flat upper and lower rails are used as anomaly detection standards to compare and determine whether the current data point is within the range of the corresponding nonwoven fabric flat upper and lower rails. At the same time, based on the tension uniformity threshold of the corresponding roll diameter slice sample group, the normalized tension data at the current moment is compared. If the current data point is within the range of the nonwoven fabric flat upper and lower rails and the tension data is less than the tension uniformity threshold, the current moment is directly determined to be a normal working condition; if... If the previous data point is within the flat upper and lower rail range of the nonwoven fabric but the tension data is greater than or equal to the tension uniformity threshold, the current working condition is marked as tension unstable. When the number of consecutive occurrences of the tension unstable working condition mark reaches a preset value, the working condition at the latest moment is marked as an abnormal candidate. If the current data point is outside the flat upper and lower rail range of the nonwoven fabric, the current working condition is directly marked as an abnormal candidate. For the abnormal candidate mark, the data collected within the preset time period after the mark is marked as the abnormal data interval. When the abnormal candidate working condition mark is detected, the image detection mechanism in the rewinding equipment is triggered synchronously to collect images of the marked abnormal data interval in real time for collaborative anomaly detection.

[0015] Preferably, after the current rewinding process is completed, if a misjudgment occurs where the data range is marked as abnormal but the image detection mechanism determines that the working condition is normal, the optimization and update phase is entered. The data collected in the current rewinding process is used as historical time series data to optimize and update the tension uniformity threshold, the nonwoven fabric flatness dataset, the fitting curve under ideal working conditions, and the upper and lower rails of the nonwoven fabric flatness. Similarly, the data of the rewinding process where a misjudgment occurs is retained to continuously perform iterative optimization.

[0016] Secondly, the present invention provides a collaborative control system for spunlace nonwoven fabric rewinding equipment based on the Internet of Things (IoT), used to implement the aforementioned collaborative control method for spunlace nonwoven fabric rewinding equipment based on the IoT. The system includes: a processor, a memory, and a communication interface. Multiple sensing devices for data acquisition are deployed and installed on the unwinding mechanism side of the rewinding equipment, and an image detection device is provided on the winding mechanism side of the rewinding equipment. The processor stores computer program instructions for implementing the aforementioned collaborative control method for spunlace nonwoven fabric rewinding equipment based on the IoT. The sensing devices include a laser rangefinder for acquiring nonwoven fabric master roll diameter data, a torque sensor for acquiring unwinding mechanism torque data, and a tension roller sensor for acquiring nonwoven fabric tension data. The communication interface is communicatively connected to each sensing device and the image detection device.

[0017] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0018] This invention, through a dynamic reference between roll diameter and torque and an upper and lower rail design, can adapt to the physical characteristics of different roll diameters, accurately identify torque anomalies, and achieve dynamic adaptation of the control mechanism to roll diameter changes. Simultaneously, by real-time monitoring of the electrical data of the unwinding mechanism, abnormal operating condition intervals can be pre-marked, enabling efficient collaborative image detection mechanisms to accurately identify abnormal operating conditions. This significantly reduces the false positive and false negative rates for abnormal operating conditions, effectively improving the quality control level of spunlace nonwoven fabric rewinding production and solving the problem of unreasonable allocation of image processing resources in existing spunlace nonwoven fabric rewinding equipment control methods. Furthermore, through a dynamic iteration mechanism, automatic optimization of control parameters can be achieved to adapt to changes in nonwoven fabric specifications. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating the implementation of the collaborative control method for spunlace nonwoven fabric rewinding equipment based on the Internet of Things, according to an embodiment of the present invention.

[0020] Figure 2 This is a structural block diagram of the collaborative control system for a spunlace nonwoven fabric rewinding equipment based on the Internet of Things, according to an embodiment of the present invention. Detailed Implementation

[0021] The technical features of the present invention will be further described in detail below with reference to the accompanying drawings, so that those skilled in the art can understand them.

[0022] A collaborative control method for spunlace nonwoven fabric rewinding equipment based on the Internet of Things, the implementation process of which is as follows: Figure 1 As shown, the specific implementation steps are as follows:

[0023] Step S1: Obtain historical time-series data of multiple rewinding processes and perform preprocessing, extract historical time-series data under normal working conditions, and obtain multiple three-dimensional data sequence sets of roll diameter-torque-tension.

[0024] In the specific production of spunlace nonwoven fabric rewinding, the unwinding mechanism in the rewinding equipment continuously unwinds, and the diameter of the master roll decreases in real time, that is, the weight of the master roll decreases. While maintaining uniform tension to maintain the normal working condition of the nonwoven fabric, the reference torque required by the unwinding mechanism will also change according to physical principles such as rotational inertia. Therefore, there is a correlation between torque and roll diameter, and data needs to be collected synchronously in real time.

[0025] Specifically, the sensor devices are deployed to collect the roll diameter data sequence, the torque data sequence of the unwinding mechanism, and the tension data sequence of the nonwoven fabric in real time. The three-dimensional data of multiple complete nonwoven fabric rewinding processes are obtained as historical time-series data. All historical time-series data are normalized. Based on the original anomaly detection method, the normalized historical time-series data are divided into normal working conditions or abnormal working conditions. The normalized historical time-series data under normal working conditions in each rewinding process is extracted to obtain the three-dimensional data sequence set of roll diameter-torque-tension in each rewinding process.

[0026] The automated rewinding equipment is equipped with torque sensors, tension sensors, and laser rangefinders, which collect data synchronously in real time at a frequency of 10Hz. The number of complete nonwoven fabric rewinding processes collected is no less than 10; the more data collected, the more accurate the subsequent calculations. The laser rangefinder acquires the roll diameter data of the nonwoven fabric master roll, the torque sensor acquires the torque data of the unwinding mechanism, and the tension sensor, such as a tension roller, acquires the tension data of the nonwoven fabric. Based on the above historical time-series data, the maximum and minimum values ​​of the corresponding data are normalized using a Min-Max normalization method to eliminate dimensions. Normal operating conditions are defined as the production process during rewinding where the nonwoven fabric does not exhibit deviations, wrinkles, or other abnormalities; conversely, abnormal operating conditions are defined as the production process where the nonwoven fabric exhibits deviations, wrinkles, or other abnormalities. Based on the existing anomaly detection method, i.e., the image detection mechanism, combined with manual inspection, historical time-series data under normal operating conditions can be distinguished and extracted.

[0027] Step S2: Construct a roll diameter window mechanism to divide multiple three-dimensional data sequence sets into several roll diameter slice sample groups, determine the tension uniformity threshold to filter the roll diameter slice samples, and obtain the nonwoven fabric flatness dataset.

[0028] Specifically, based on the quality parameters of spunlace nonwoven fabric and the operating speed of the rewinding equipment in actual application scenarios, the roll diameter window reference length is set. The normalized roll diameter range is divided into multiple equal-length roll diameter intervals. The three-dimensional data sequence set for each rewinding process is divided according to the roll diameter window reference length, resulting in several roll diameter slice samples corresponding to each rewinding process. Roll diameter slice samples from the same roll diameter interval in each rewinding process are grouped into a roll diameter slice sample group, resulting in several roll diameter slice sample groups. The number of roll diameter slice sample groups is equal to the reciprocal of the roll diameter window reference length. Tension data sequences are extracted from each roll diameter slice sample within each roll diameter slice sample group, and the interquartile range method is used to calculate the tension data. The interquartile range of the data sequence is used as the tension fluctuation of the corresponding roll diameter slice sample. The tension fluctuation of each roll diameter slice sample within each roll diameter slice sample group is obtained. The smaller the tension fluctuation value, the more uniform the tension. Based on the tension fluctuation, a histogram corresponding to each roll diameter slice sample group is constructed. The Otsu's method is used to determine the tension uniformity threshold of each roll diameter slice sample group. Roll diameter slice samples within each roll diameter slice sample group with tension fluctuation less than the corresponding tension uniformity threshold are taken as tension uniform samples. The corresponding roll diameter-torque data pairs in the tension uniform samples are taken as data points. The set of all data points in all tension uniform samples is taken as the nonwoven fabric flattening dataset.

[0029] The normalized roll diameter data ranges from 0 to 1, and the roll diameter window reference length can be set to 0.01, resulting in 100 roll diameter intervals. This divides the 3D data sequence set of each rewinding process into 100 roll diameter slice samples, and then divides all roll diameter slice samples into 100 roll diameter slice sample groups. During the rewinding process, the length of the data sequence extracted from different roll diameter intervals varies, resulting in differences in the amount of data contained. The larger the diameter of the master roll, the smaller the rate of change of the roll diameter, meaning that the roll diameter interval corresponding to a large roll diameter contains more sampled data. However, since the master roll with a large roll diameter is heavier and has stronger rotational inertia, it is less affected by the torque fluctuation of the unwinding mechanism and has lower sensitivity to torque fluctuations. Therefore, the roll diameter interval of a large roll diameter does not need to be divided into overly fine data. More data is needed to reflect the relationship between the roll diameter and torque within the roll diameter interval, which is consistent with the rule of extracting roll diameter slice samples from the roll diameter interval.

[0030] Step S3: Based on the nonwoven fabric flattening dataset, calculate the sample weights of each sample in the dataset and perform nonlinear fitting to obtain the fitting curve under ideal working conditions.

[0031] Specifically, the tension fluctuation of each roll diameter slice sample within each roll diameter slice sample group is extracted. Based on the nonwoven fabric flatness dataset, the tension fluctuation of each tension-uniform sample within each roll diameter slice sample group is extracted. The mean of the tension fluctuation of each roll diameter slice sample within each roll diameter slice sample group is used as the baseline fluctuation value of the corresponding roll diameter slice sample group. The ratio between the tension fluctuation of each tension-uniform sample within each roll diameter slice sample group and the corresponding baseline fluctuation value is calculated. The natural exponential function is used to perform inverse mapping of the comparison value, and the mapped value is used as the sample weight of the corresponding tension-uniform sample to obtain the sample weight of each tension-uniform sample within each roll diameter slice sample group. Mature nonlinear fitting tools in the field of industrial control, such as the data fitting function lsqcurvefit in the digital modeling software Matlab, are directly called. The nonwoven fabric flatness dataset is used as input. During the iterative optimization process of the fitting curve function, the sample weight of each tension-uniform sample is introduced to calculate the weighted loss until the fitting curve function with the minimum weighted loss is obtained as the fitting curve under ideal working conditions.

[0032] In this study, the normalized roll diameter data from the nonwoven fabric flattening dataset is used as the independent variable, and the normalized torque data is used as the dependent variable. A fitting curve function of roll diameter-torque can be obtained through a fitting tool. During the iterative optimization of the fitting curve function, it is necessary to calculate the distance of each data point in the nonwoven fabric flattening dataset from the fitting curve as the fitting loss. The sample weights related to tension fluctuation are introduced to calculate the weighted loss. This is to assign higher fitting weights to samples with more uniform tension, thereby forcing the fitting curve to move closer to the ideal working condition with uniform tension. This ensures that the fitting curve can better fit the relationship between roll diameter and torque under the ideal working condition. Finally, the fitting curve function with the minimum weighted loss is output as the fitting curve under the ideal working condition, which serves as the basis for the subsequent construction of the middle rail of the upper and lower rails of the nonwoven fabric flattening.

[0033] Step S4: Based on the fitting curve under ideal working conditions, construct the flat upper and lower rails of the nonwoven fabric.

[0034] Based on the nonwoven fabric flattening dataset, the roll diameter-torque fitting curve for the entire roll diameter range under ideal working conditions was obtained. By dividing the roll diameter range based on the roll diameter window reference length, torque reference curves under ideal working conditions for different roll diameter ranges and roll diameter slice sample groups can be obtained. Then, using the torque reference curve under ideal working conditions as the center line, the deviation of each data point in the nonwoven fabric flattening dataset from the torque reference curve is quantified to select the optimal radius threshold and filter out the core data points in the nonwoven fabric flattening dataset. Based on the core data points, nonwoven fabric flattening upper and lower rails that dynamically change with roll diameter are constructed to provide a quantitative boundary for judging torque anomalies in real-time production.

[0035] Specifically, the absolute value of the vertical distance from the data point to the fitted curve under ideal working conditions is taken as the torque deviation distance of the corresponding data point. The torque deviation distance of each data point in the nonwoven fabric flattening dataset is calculated to obtain a set of deviation distances. Statistical calculations are performed on the set of deviation distances to obtain the median and interquartile range values. A radius threshold is set, with its upper limit being the sum of the median and interquartile range values ​​of the deviation distance set and its lower limit being the median value. The upper limit of the radius threshold is used as the initial value, and a step size for gradually decreasing the radius threshold is set based on the upper limit of the radius threshold. The set of values ​​for the radius threshold is obtained by taking the lower limit of the value; the interquartile range of the deviation distance set of 0.1 times is used as the threshold neighborhood coefficient, the sum of the radius threshold and the threshold neighborhood coefficient is used as the upper limit of classification, and the difference between the two is used as the lower limit of classification. Based on the torque deviation distance, the data points in the non-woven fabric flattening dataset are classified and filtered. Data points with a torque deviation distance less than the lower limit of classification are classified as core data points, data points with a torque deviation distance greater than the upper limit of classification are classified as edge data points, and data points with a torque deviation distance between the upper limit of classification and the lower limit of classification are classified as fuzzy data points.

[0036] Among them, the torque deviation distance is used to quantify the degree of deviation between the data points and the fitted curve under ideal working conditions. The step size of the radius threshold can be set to 5% of the interquartile range of the deviation distance set.

[0037] Furthermore, a classification loss function is constructed, setting the classification loss function value of fuzzy data points to 1 and the classification loss function value of core data points and edge data points to 0. Based on a certain value within the radius threshold set, the fuzzy data points under that value are statistically analyzed. Based on the sample weights of the corresponding samples in the nonwoven fabric flattening dataset, a corresponding sample weight is matched for each fuzzy data point under that value. The cumulative value of the product of the classification loss function value of all fuzzy data points under that value and the corresponding sample weight is used as the weighted classification loss value corresponding to that value. The set of radius threshold values ​​is traversed to obtain the weighted classification loss value corresponding to each radius threshold value in the set. The radius threshold value corresponding to the minimum value among all weighted classification loss values ​​is used as the optimal radius threshold, thus obtaining the corresponding set of core data points.

[0038] The purpose of constructing a classification loss function and setting it to 0 or 1 is to simplify classification judgment and clarify the classification boundary, highlighting the loss of classification results due to fuzzy data points. The introduction of sample weights is to highlight the classification priority of high-quality data points. High-quality data points with more uniform tension and smaller torque deviation reflect a more realistic diameter-torque relationship. Giving high-quality data points higher weights can make them have a greater impact on the calculation results, highlighting their dominant role. When selecting the optimal radius threshold, it is necessary to ensure that the overall torque deviation of the data points is small while taking into account the uniformity and stability of the tension of the data points. This ensures that the core data points meet the ideal working condition requirements of fitting the torque benchmark and uniform tension, thereby forcing the classification results to converge towards the area where high-quality data points are concentrated. The core data points selected based on the optimal radius threshold are all ideal data with uniform tension and fitting the fitted curve, thus providing a highly reliable data foundation for the subsequent construction of nonwoven fabric flat upper and lower rails.

[0039] Furthermore, based on the roll diameter range of each roll diameter slice sample group, the core data points are categorized, and the torque data of each core data point within each roll diameter slice sample group is extracted. The torque standard deviation of the core data points within each roll diameter slice sample group is calculated. Based on the three sigma law, the sum of the fitted curve function under ideal working conditions and three times the torque standard deviation of a certain roll diameter slice sample group is taken as the nonwoven fabric flattening upper rail corresponding to that roll diameter slice sample group. The difference between the fitted curve function under ideal working conditions and three times the torque standard deviation of a certain roll diameter slice sample group is taken as the nonwoven fabric flattening lower rail corresponding to that roll diameter slice sample group. Similarly, the nonwoven fabric flattening upper and lower rails corresponding to each roll diameter slice sample group are obtained.

[0040] This method uses the fitted curve under ideal working conditions as a benchmark to ensure that the upper and lower rails always fluctuate around the benchmark trend of ideal working conditions, avoiding deviations from physical laws at the boundaries. The upper and lower rails for nonwoven fabric flattening are determined based on slice sample groups of each roll diameter. The purpose is to adaptively and dynamically adjust the bandwidth of the upper and lower rails for nonwoven fabric flattening according to changes in roll diameter. Based on the physical characteristic that torque fluctuations are more unstable at larger roll diameters, the range of upper and lower rails for nonwoven fabric flattening is wider when the roll diameter is larger. The torque standard deviation obtained based on core data points only reflects the natural fluctuation range of torque under normal working conditions, eliminating the interference of edge data points that may be noise and data points with large tension fluctuations. This ensures that the boundaries of the upper and lower rails can accurately wrap the torque fluctuations under normal working conditions, forming a normal working condition data range of roll diameter-torque. Compared with the fixed threshold scheme in traditional methods that ignore the influence of roll diameter, this method can effectively improve the identification accuracy of abnormal working condition data and reduce the false positive rate and false negative rate.

[0041] Step S5: Based on the real-time marking of abnormal data intervals in the current rewinding process using the nonwoven fabric flat upper and lower rails, and coordinate with the image detection mechanism in the rewinding equipment to determine the abnormality.

[0042] Using the flat upper and lower rails of the nonwoven fabric corresponding to each roll diameter range as the anomaly detection standard, a three-level detection mechanism of real-time data monitoring, anomaly candidate marking, and image verification confirmation is used to accurately identify the abnormal working conditions that occur during the nonwoven fabric rewinding process.

[0043] Specifically, the system synchronously collects and normalizes the roll diameter data of the nonwoven fabric master roll, the torque data of the unwinding mechanism, and the tension data of the nonwoven fabric during the current rewinding process. The normalized roll diameter-torque data pair is used as the current data point. Based on the normalized roll diameter data at the current moment, the current data point is classified into the corresponding roll diameter slice sample group. The corresponding nonwoven fabric flat upper and lower rails are used as anomaly detection standards to compare and determine whether the current data point is within the range of the corresponding nonwoven fabric flat upper and lower rails. At the same time, the normalized tension data at the current moment is compared based on the tension uniformity threshold of the corresponding roll diameter slice sample group.

[0044] In the process of normalizing real-time data, the Min-Max normalization method in step S1 is also used, and the maximum and minimum values ​​in the historical time series data are also used.

[0045] If the current data point is within the flat upper and lower rails of the nonwoven fabric and the tension data is less than the tension uniformity threshold, then the current moment is directly determined to be a normal working condition. If the current data point is within the flat upper and lower rails of the nonwoven fabric but the tension data is greater than or equal to the tension uniformity threshold, the current working condition is marked as tension unstable. When the number of consecutive occurrences of the tension unstable working condition mark reaches a preset value, the working condition at the latest moment is marked as an abnormal candidate. The preset value for the number of consecutive occurrences of the tension unstable working condition mark can be set to 10 times based on the data sampling frequency. If the current data point is outside the flat upper and lower rails of the nonwoven fabric, then the current working condition is directly marked as an abnormal candidate.

[0046] Specifically, for anomaly candidate markers, data collected within a preset time period after the marker time is marked as anomaly data interval. When an anomaly candidate is detected, the image detection mechanism in the rewinding device is simultaneously triggered to collect images of the marked anomaly data interval in real time for collaborative anomaly detection. When the image detection mechanism determines that an anomaly has been detected, an alarm is immediately triggered and the equipment is shut down or corrected. If the image detection mechanism does not detect an anomaly within the anomaly data interval, it determines whether the condition has returned to normal based on the real-time data at the last moment of the anomaly data interval. If the condition has returned to normal, the image detection mechanism stops; if the condition has not returned to normal, the image detection continues for the next anomaly data interval. By marking anomaly candidates for data early warning and then efficiently collaborating with the image detection mechanism for identification and verification, the dual judgment ensures the accuracy of anomaly determination and effectively reduces the false positive rate and false negative rate.

[0047] Step S6: After the current rewinding process is completed, optimize and update the range of the upper and lower rails for flattening the nonwoven fabric.

[0048] Specifically, after the current rewinding process is completed, if a misjudgment occurs where the data range is marked as abnormal but the image detection mechanism determines that the working condition is normal, the optimization and update phase will begin. The data collected from the current rewinding process will be used as historical time-series data to optimize and update the tension uniformity threshold, the nonwoven fabric flatness dataset, the fitting curve under ideal working conditions, and the upper and lower rails of the nonwoven fabric flatness. Similarly, the data from each rewinding process where a misjudgment occurs will be retained to continuously perform iterative optimization and achieve a closed loop of pre-labeling, image verification, and control optimization.

[0049] In addition, sampling strategies can be combined to perform image verification on data judged as normal working conditions in order to verify the accuracy of the normal working condition judgment. While ensuring production safety, this facilitates subsequent feedback and optimization of the anomaly detection mechanism. If a misjudgment occurs after image verification of normal working conditions, the upper and lower rails of the nonwoven fabric in the corresponding roll diameter range will be shrunken and adjusted when entering the optimization and update stage.

[0050] This invention also discloses a collaborative control system for a spunlace nonwoven fabric rewinding equipment based on the Internet of Things, the structure of which is as follows: Figure 2 As shown, the collaborative control method for the above-mentioned IoT-based spunlace nonwoven fabric rewinding equipment includes: a processor, a memory, and a communication interface. Multiple sensors for data acquisition are deployed on the unwinding mechanism side of the rewinding equipment, and an image detection device is provided on the winding mechanism side. The processor stores computer program instructions for implementing the above-mentioned IoT-based collaborative control method for the spunlace nonwoven fabric rewinding equipment. The sensors include a laser rangefinder for acquiring nonwoven fabric roll diameter data, a torque sensor for acquiring unwinding mechanism torque data, and a tension roller sensor for acquiring nonwoven fabric tension data. The communication interface is connected to each sensor and the image detection device. The purpose of the communication interface and the image detection device is to trigger the image detection process by issuing a control signal from the processor when abnormal data occurs, thereby achieving efficient collaborative detection of abnormal working conditions, reducing the false judgment rate of nonwoven fabric deviation anomalies, improving the quality control level of spunlace nonwoven fabric rewinding production, and simultaneously solving the problem of unreasonable allocation of image processing resources in existing rewinding equipment control methods.

[0051] The embodiments included in this invention are merely descriptions of preferred embodiments of the invention and are not limited to the precise structures described above and shown in the accompanying drawings. Various modifications and changes can be made without departing from the scope of protection. Any variations and improvements made by those skilled in the art to the technical solutions of this invention without departing from the design concept of this invention should fall within the scope of protection of this invention.

Claims

1. A method for collaborative control of a water-jet non-woven fabric rewinding device based on the Internet of Things, characterized in that: The historical time series data of multiple rewinding processes are acquired and preprocessed, the historical time series data under normal working conditions are extracted, and a plurality of three-dimensional data sequence sets of winding diameter-torque-tension are obtained; a winding diameter window mechanism is constructed to divide the plurality of three-dimensional data sequence sets into a plurality of winding diameter slice sample groups, a tension uniformity threshold is determined to screen the winding diameter slice samples, and a non-woven fabric flatness data set is obtained; Based on the non-woven fabric flatness data set, the sample weights of each sample in the data set are calculated and nonlinear fitting is performed to obtain a fitting curve under ideal working conditions; based on the fitting curve under ideal working conditions, non-woven fabric flatness upper and lower tracks are constructed; based on the non-woven fabric flatness upper and lower tracks, abnormal data intervals in the current rewinding process are marked in real time, and abnormality determination is performed in cooperation with the image detection mechanism in the rewinding equipment; after the current rewinding process is completed, the range interval of the non-woven fabric flatness upper and lower tracks is optimized and updated; The sensing device is deployed to synchronously collect winding diameter data sequences of the non-woven fabric parent roll, torque data sequences of the unwinding mechanism, and tension data sequences of the non-woven fabric, acquire three-dimensional data of multiple complete non-woven fabric rewinding processes as historical time series data, and normalize all historical time series data; based on the original abnormality detection method, the normalized historical time series data are divided into normal working conditions or abnormal working conditions, the normalized historical time series data under normal working conditions in each rewinding process are extracted, and three-dimensional data sequence sets of winding diameter-torque-tension in each rewinding process are obtained; Based on the tension uniformity threshold, the non-woven fabric flatness data set is screened to obtain a fitting curve under ideal working conditions, the fitting curve is a curve function between winding diameter and torque, winding diameter is the independent variable, and torque is the dependent variable; the non-woven fabric flatness data set is further screened to obtain a set of core data points, the torque standard deviation of each winding diameter slice sample group is calculated, and based on the fitting curve under ideal working conditions and the three-sigma law, the non-woven fabric flatness upper and lower tracks corresponding to each winding diameter slice sample group are obtained as the abnormality detection standard.

2. The method according to claim 1, wherein, The construction of the winding diameter window mechanism divides a plurality of three-dimensional data sequence sets into a plurality of winding diameter slice sample groups, determines a tension uniformity threshold value to screen the winding diameter slice samples, and obtains a non-woven fabric flatness data set, including: setting a winding diameter window reference length according to quality parameters of spunlace non-woven fabric in an actual application scenario and a running speed of a rewinding device, dividing a normalized winding diameter value range into a plurality of equal-length winding diameter intervals, dividing the three-dimensional data sequence set of each rewinding process according to the winding diameter window reference length, obtaining a plurality of winding diameter slice samples corresponding to each rewinding process, taking the winding diameter slice samples in the same winding diameter interval in each rewinding process as a winding diameter slice sample group, obtaining a plurality of winding diameter slice sample groups, and the number of the winding diameter slice sample groups is equal to the inverse of the winding diameter window reference length; extracting tension data sequences in each winding diameter slice sample in each winding diameter slice sample group, using the quartile range method to calculate the quartile range value of each tension data sequence as the tension fluctuation degree of the corresponding winding diameter slice sample, obtaining the tension fluctuation degree of each winding diameter slice sample in each winding diameter slice sample group; based on the tension fluctuation degree, a histogram corresponding to each winding diameter slice sample group is constructed, and the maximum inter-class variance method is used to determine the tension uniformity threshold value of each winding diameter slice sample group, the winding diameter slice samples with a tension fluctuation degree less than the corresponding tension uniformity threshold value in each winding diameter slice sample group are taken as tension uniformity samples, the corresponding winding diameter-torque data pairs in the tension uniformity samples are taken as data points, and the set of all data points in all tension uniformity samples is taken as the non-woven fabric flatness data set.

3. The method according to claim 2, wherein, Based on the non-woven fabric flatness data set, the sample weight of each sample in the data set is calculated and nonlinear fitting is performed to obtain a fitting curve under an ideal working condition, including: extracting the tension fluctuation degree of each winding diameter slice sample in each winding diameter slice sample group, and based on the non-woven fabric flatness data set, extracting the tension fluctuation degree of each tension uniformity sample in each winding diameter slice sample group, taking the mean value of the tension fluctuation degrees of each winding diameter slice sample in each winding diameter slice sample group as the reference fluctuation value of the corresponding winding diameter slice sample group, calculating the ratio between the tension fluctuation degree of each tension uniformity sample in each winding diameter slice sample group and the corresponding reference fluctuation value, using a natural exponential function to inversely map the ratio, and taking the mapped value as the sample weight of the corresponding tension uniformity sample to obtain the sample weight of each tension uniformity sample in each winding diameter slice sample group; directly calling a mature nonlinear fitting tool in the industrial control field, taking the non-woven fabric flatness data set as input, introducing the sample weight of each tension uniformity sample to calculate a weighted loss in the iterative optimization process of the fitting curve function, and obtaining a fitting curve function with the minimum weighted loss as the fitting curve under the ideal working condition.

4. The method according to claim 2, wherein, The fitting curve under the ideal working condition is used to construct the upper and lower rails of the non-woven fabric smoothing, including: taking the absolute value of the perpendicular distance from the data point to the fitting curve under the ideal working condition as the torque deviation distance of the corresponding data point, calculating the torque deviation distance of each data point in the non-woven fabric smoothing data set to obtain a deviation distance set, performing statistical calculation on the deviation distance set to obtain the median and interquartile range value of the deviation distance set; setting a radius threshold, the upper limit of the value of which is the sum of the median and the interquartile range value of the deviation distance set, and the lower limit of the value of which is the value of the median, taking the upper limit of the value of the radius threshold as the initial value, setting a step length for gradually decreasing the radius threshold, and obtaining a value set of the radius threshold based on the upper limit and the lower limit of the value of the radius threshold; taking 0.1 times the interquartile range value of the deviation distance set as a threshold neighborhood coefficient, taking the sum of the radius threshold and the threshold neighborhood coefficient as the upper limit of classification, taking the difference between the two as the lower limit of classification, and classifying and screening the data points in the non-woven fabric smoothing data set based on the torque deviation distance, dividing the data points with a torque deviation distance less than the lower limit of classification into core data points, dividing the data points with a torque deviation distance greater than the upper limit of classification into edge data points, and dividing the data points with a torque deviation distance between the upper limit and the lower limit of classification into fuzzy data points.

5. The method according to claim 4, wherein, The fitting curve under the ideal working condition is used to construct the upper and lower rails of the non-woven fabric smoothing, and further includes: constructing a classification loss function, setting the classification loss function value of the fuzzy data points to 1, and setting the classification loss function value of the core data points and the edge data points to 0; based on a certain value in the value set of the radius threshold, counting the fuzzy data points under the value, matching a corresponding sample weight for each fuzzy data point under the value based on the sample weight of each sample in the non-woven fabric smoothing data set, taking the cumulative value of the product of the classification loss function value of all fuzzy data points under the value and the corresponding sample weight as the weighted classification loss value corresponding to the value; traversing the value set of the radius threshold to obtain the weighted classification loss value corresponding to each radius threshold value in the set, taking the radius threshold value corresponding to the minimum value in all weighted classification loss values as the optimal radius threshold, and obtaining the corresponding set of core data points.

6. The method according to claim 5, wherein, The fitting curve under the ideal working condition is used to construct the upper and lower rails of the non-woven fabric smoothing, and further includes: classifying the core data points based on the roll diameter interval of each roll diameter slice sample group, extracting the torque data of each core data point in each roll diameter slice sample group, and calculating the torque standard deviation of the core data points in each roll diameter slice sample group; based on the three-sigma law, taking the sum of the fitting curve function under the ideal working condition and three times the torque standard deviation of a certain roll diameter slice sample group as the upper rail of the non-woven fabric smoothing corresponding to the roll diameter slice sample group, and taking the difference between the fitting curve function under the ideal working condition and three times the torque standard deviation of a certain roll diameter slice sample group as the lower rail of the non-woven fabric smoothing corresponding to the roll diameter slice sample group, and similarly obtaining the upper and lower rails of the non-woven fabric smoothing corresponding to each roll diameter slice sample group.

7. The Internet of Things-based coordinated control method for the spunlace nonwoven fabric rewinding equipment according to any one of claims 2 to 6, characterized in that, The abnormal data interval in the current rewinding process is marked in real time based on the non-woven fabric flat upper and lower rails, and the image detection mechanism in the rewinding equipment is cooperated for abnormality determination, including: three-dimensional data of the current rewinding process is collected in real time and normalized, and the normalized roll diameter-torque data pair is taken as the current data point, the current data point is classified into the corresponding roll diameter slice sample group based on the normalized roll diameter data at the current time, and the corresponding non-woven fabric flat upper and lower rails are taken as the abnormality detection standard, and whether the current data point is located within the range of the corresponding non-woven fabric flat upper and lower rails is compared and judged, and the normalized tension data at the current time is compared based on the tension uniformity threshold of the corresponding roll diameter slice sample group; if the current data point is located within the range of the non-woven fabric flat upper and lower rails and the tension data is less than the tension uniformity threshold, it is directly determined that the current time is a normal working condition; if the current data point is located within the range of the non-woven fabric flat upper and lower rails but the tension data is greater than or equal to the tension uniformity threshold, the working condition at the current time is marked as tension instability, and when the number of continuous occurrence of the working condition marked as tension instability reaches a preset value, the working condition at the latest time is marked as an abnormal candidate; if the current data point is located outside the range of the non-woven fabric flat upper and lower rails, the working condition at the current time is directly marked as an abnormal candidate; for the abnormal candidate mark, the data collected in the preset period after the marking time is marked as an abnormal data interval; when the working condition marked as an abnormal candidate is detected, the image detection mechanism in the rewinding equipment is synchronously triggered to collect images of the marked abnormal data interval in real time for cooperative abnormality inspection. 8.The method according to claim 7, wherein, After the current rewinding process is completed, the range interval of the non-woven fabric flat upper and lower rails is optimized and updated, including: after the current rewinding process is completed, if the misjudgment condition that the abnormal data interval is marked but the image detection mechanism determines that the working condition is normal occurs, the optimization and update stage is entered, the data of the current rewinding process collected is taken as historical time sequence data, the tension uniformity threshold, the non-woven fabric flat data set, the fitting curve under the ideal working condition and the non-woven fabric flat upper and lower rails are optimized and updated, and the data of the rewinding process in each misjudgment condition is retained for continuous iterative optimization.

9. The water-jet non-woven fabric rewinding equipment collaborative control system based on the Internet of Things, characterized in that: The rewinding equipment includes a processor, a memory, and a communication interface, a plurality of sensing devices for collecting data are installed on one side of a unwinding mechanism of the rewinding equipment, an image detection device is arranged on one side of a winding mechanism of the rewinding equipment, the processor stores computer program instructions for realizing the Internet of Things-based spunlace non-woven fabric rewinding equipment cooperative control method of any one of claims 1 to 8, the sensing devices include a laser ranging sensor for collecting non-woven fabric parent roll diameter data, a torque sensor for collecting unwinding mechanism torque data, and a tension roller sensor for collecting non-woven fabric tension data, and the communication interface is in communication connection with each of the sensing devices and the image detection device.

Images